The previous code calculates each step time via an "exponential
search" followed by a "false position with Illinois algorithm" search.
Replace with a "secant method" with "bounds check" search. This
simplifies the code, improves the performance, and does a better job
of finding steps near a direction change.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
When in debug output mode, make sure to register the fd correctly.
Otherwise, the poll() call will use the 0 fd which could cause random
results.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Avoid using the absolute E position when calculating pressure advance
as that position can grow arbitrarily large, which can result in
"numerical stability" problems. That instability could eventually
lead to internal errors during step compression.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
The default on X86 32bit machines is to use 80bit floating point math
(as found in the ancient "387 coprocessor"). This can cause numerical
stability problems. Set the compiler flags to make sure the newer SSE
math is always used on X86.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
The optimized version is only available if it's the only axis used.
This fixes input_shaper on corexz.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Combine the shaper pulses and count of those pulses into a `struct
shaper_pulses`. Allocate space for the maximum number of pulses in
`struct input_shaper`.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Use a more strict check for determining if the iterative solver has
correctly found a step - the guess must be within 1 picometer of the
target or correct to within 1 nanosecond.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
This prevents some cases where the iterative solver fails to converge
in a reasonable time, causing "no next step" errors.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
For example, Raspbian GNU/Linux 8 (jessie) uses an old GCC version 4.9.2
which uses -std=gnu90 by default.
Signed-off-by: Dmitry Butyugin <dmbutyugin@google.com>
Input shaping can help to reduce printer vibrations due to resonances
and eliminate or reduce ghosting in prints.
Signed-off-by: Dmitry Butyugin <dmbutyugin@google.com>
Avoid using doubles as flags when generating steps during "lead up to
stepper activity". Instead, explicitly track how many inactive move
entries are skipped and only generate steps for those entries.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Don't assume the times passed to itersolve_gen_steps_range() are
contained within the move. It's possible the checks in
itersolve_generate_steps() may round to different values when
converting the times to a relative move time.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Some stepper motor drivers do not respond well to rapid "step +
direction change + step" events. In particular, it is believed this
can cause "over current" events on the tmc2208 drivers when they are
in "stealthchop" mode. Detect these events and remove them from the
generated step times.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Some stepper motor drivers do not respond well to rapid "step +
direction change + step" events. In particular, it is believed this
can cause "over current" events on the tmc2208 drivers when they are
in "stealthchop" mode. Detect these events and remove them from the
generated step times.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
If the stepper changes direction then the average velocity since the
last step pulse isn't a good indicator of the next step pulse.
Instead, reset the bounds checking to use a low starting guess.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
The queue_append system predates the iterative solver - it was useful
when many different kinematic functions directly added steps to the
step compression queues. With the iterative solver being the only
source of step generation, it is simpler to directly add steps from
the iterative solver code.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Add ability for the host code to get a notification when the ack for a
command sent to the micro-controller is received. This is in
preparation for improved detection of message loss between mcu and
host.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
In the rare case where the step generation code finds an interval of
zero, it was possible for an integer division of a negative number to
cause an incorrect result. Fix that by using the idiv_up() and
idiv_down() helpers.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
If a null move is added as the first move on the trapezoid motion
queue then it could have a very large move->move_t value. This could
cause numerical stability issues with some advanced kinematic
formulas. Place a limit on the move_t value to avoid this.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Implement the pressure advance calculations while performing the
definitive integral calculations. This simplifies both the
extruder.py and kin_extruder.c code.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Support averaging the extruder position over a time range to "smooth
out" the velocity changes that occur during pressure advance.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Support calculating the definitive integral of a cartesian axis
position over a time range of the movement queue.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Add support for generating steps from kinematic functions that
calculate step times based on a range of the motion queue. It
requires scanning for step generation during the lead up to stepper
activity (when the stepper would nominally be idle). And it requires
scanning for step generation just after a stepper has nominally become
idle.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Use sentinels to make list traversal code simpler. Also add in null
moves so that there are no time gaps in the list.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Now that the only caller to set_commanded_position() is
set_position(), remove the set_commanded_position() method and pass
the set_position() call directly to the itersolve.c code.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Minor rework of the loop to reduce the chance that numerical stability
could cause an infinite loop.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Calculate the ratio of axis distance to total move distance (axis_d /
move_d) and store in a new member variable axes_r. This avoids
needing to recalculate the value in other code.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Only track a single acceleration movement in a 'struct move' instance.
Break the classic trapezoid movement (accel, cruise, decel) into three
separate movements. This simplifies the calculation logic.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Now that all callers use the trapq system to queue moves, it is no
longer necessary to individually allocate and fill a 'struct move'.
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Kevin O'Connor